Merge branch 'master' of github.com:chainer/chainer into add-try-trai…

…ner-class-tutorial

appveyor.yml

@@ -64,9 +64,9 @@ build_script:
 
 test_script:
   # Run the project tests
-  - "%CMD_IN_ENV% pip install nose mock hacking"
+  - "%CMD_IN_ENV% pip install mock hacking"
+  - "%CMD_IN_ENV% pip install nose pytest pytest-timeout pytest-cov"
   - "flake8"
-  - "cd tests\\chainer_tests"
   # Avoid interuption confirmation of cmd.exe
-  - "echo nosetests -a \"!gpu,!slow\" > tmp.bat"
+  - "echo python -m pytest --timeout=60 -m \"not gpu and not multi_gpu and not cudnn and not slow\" tests > tmp.bat"
   - "call tmp.bat < nul"

chainer/function.py

@@ -135,7 +135,7 @@ def backward(self, target_input_indexes, grad_outputs):
             if gxs[i] is None:
                 g = None
             else:
-                # Intentionallly not passing requires_grad=False so that
+                # Intentionally not passing requires_grad=False so that
                 # backprop routines can raise an error when a further backprop
                 # is attempted against this gradient variable.
                 g = variable.Variable(gxs[i])
@@ -487,7 +487,7 @@ def retain_inputs(self, indexes):
 
         Args:
             indexes (iterable of int): Indexes of input variables that the
-                function does not require for backprop.
+                function will require for backprop.
 
         """
         self.node.retain_inputs(indexes)
@@ -513,7 +513,7 @@ def retain_outputs(self, indexes, retain_after_backward=False):
 
         Args:
             indexes (iterable of int): Indexes of input variables that the
-                function does not require for backprop.
+                function will require for backprop.
 
             retain_after_backward (bool): This option has no effect. It is
                 left only for the backward compatibility.

chainer/functions/activation/relu.py

@@ -58,7 +58,7 @@ class ReLUGrad2(function_node.FunctionNode):
 
     This function takes 2 variables b and c, and
     computes f(b, c) = sign(b) * c with backpropagation
-    where operations are dones in elementwise manner
+    where operations are done in elementwise manner
     and sign(x) = 1 when x > 0 is positive and 0 otherwise.
 
     As the gradient of f with respect to b is 0,

chainer/utils/__init__.py

@@ -4,6 +4,8 @@
 
 
 # import class and function
+from chainer.utils.conv import get_conv_outsize  # NOQA
+from chainer.utils.conv import get_deconv_outsize  # NOQA
 from chainer.utils.experimental import experimental  # NOQA
 from chainer.utils.walker_alias import WalkerAlias  # NOQA
 

chainer/utils/conv.py

@@ -5,6 +5,27 @@
 
 
 def get_conv_outsize(size, k, s, p, cover_all=False, d=1):
+    """Calculates output size of convolution.
+
+    This function takes the size of input feature map, kernel, stride, and
+    pooling of one particular dimension, then calculates the output feature
+    map size of that dimension.
+
+    .. seealso:: :func:`~chainer.utils.get_deconv_outsize`
+
+    Args:
+        size (int): The size of input feature map. It usually is the length of
+            a side of feature map.
+        k (int): The size of convolution kernel.
+        s (int): The size of stride.
+        p (int): The size of padding.
+        cover_all (bool): Use ``cover_all`` option or not.
+        d (int): The size of dilation.
+
+    Returns:
+        int: The expected output size of the convolution operation.
+
+    """
     dk = k + (k - 1) * (d - 1)
     if cover_all:
         return (size + p * 2 - dk + s - 1) // s + 1
@@ -13,6 +34,26 @@ def get_conv_outsize(size, k, s, p, cover_all=False, d=1):
 
 
 def get_deconv_outsize(size, k, s, p, cover_all=False):
+    """Calculates output size of deconvolution.
+
+    This function takes the size of input feature map, kernel, stride, and
+    pooling of one particular dimension, then calculates the output feature
+    map size of that dimension.
+
+    .. seealso:: :func:`~chainer.utils.get_conv_outsize`
+
+    Args:
+        size (int): The size of input feature map. It usually is the length of
+            a side of feature map.
+        k (int): The size of deconvolution kernel.
+        s (int): The size of stride.
+        p (int): The size of padding.
+        cover_all (bool): Use ``cover_all`` option or not.
+
+    Returns:
+        int: The expected output size of the deconvolution operation.
+
+    """
     if cover_all:
         return s * (size - 1) + k - s + 1 - 2 * p
     else:

docs/source/reference/util.rst

@@ -4,6 +4,7 @@ Utilities
 .. toctree::
    :maxdepth: 2
 
+   util/conv
    util/cuda
    util/algorithm
    util/reporter

docs/source/reference/util/conv.rst

@@ -0,0 +1,9 @@
+Convolution/Deconvolution utilities
+-----------------------------------
+
+.. autosummary::
+   :toctree: generated/
+   :nosignatures:
+
+   chainer.utils.get_conv_outsize
+   chainer.utils.get_deconv_outsize